Blind Source Separation on Convolutive Mixture Real Recording
The figures below show how a variety of our blind source separation algorithms can extract clean sources. The mixing signals were from http://www.cnl.salk.edu/~tewon. The figures below represent the following:
- Mixing input 1 of two male voices from http://www.cnl.salk.edu/~tewon.
- Mixing input 2 of two male voices.
- Output 1 from generalized anti-Hebbian learning (ICNN 1996, MWSCAS 1998, ICASSP 1999).
- Output 2 from generalized anti-Hebbian learning.
- Output 1 from the frequency method (ICA 1999).
- Output 2 from the frequency method.
- Mixing input 1 of a male voice and music from http://www.cnl.salk.edu/~tewon.
- Mixing input 2 of a male voice and music.
- Output 1 from generalized anti-Hebbian learning (ICNN 1996, MWSCAS 1998, ICASSP 1999).
- Output 2 from generalized anti-Hebbian learning.
- Output 1 from the frequency method (ICA 1999).
- Output 2 from the frequency method.
Click on any of the figures to hear the corresponding audio signals.
The simulation of speech-speech mixture
MIXING SIGNALS
OUTPUTS FROM GENERALIZED ANTI-HEBBIAN LEARNING
OUTPUTS FROM FREQUENCY METHOD
The simulation of speech-music mixture
MIXING SIGNALS
OUTPUTS FROM GENERALIZED ANTI-HEBBIAN LEARNING
OUTPUTS FROM FREQUENCY METHOD